Devoxx Morocco 2026 - Why Your RAG Agent Is Confidently Wrong: Retrieval Choices That Actually Matter
Welcome to the landing page for the workshop Why Your RAG Agent Is Confidently Wrong: Retrieval Choices That Actually Matter at Devoxx Morocco 2026.
This repo intends to provide instructions and materials for our workshop. Instruction will be provided and tasks are self-paced (with help from me should you need it). This workshop is broken down into 3 majors sections:
- Introduction to Alternative Data Retrieval Methods for RAG
- Highlight RAG Challenges When Answering User's Questions
- How Hybrid RAG Architectures Provide Answers, Not Just Semantically Similar Text Parading as Answers
Participants should ensure they have the minimum requirements:
- A Linux or Mac Developer’s Laptop with enough memory (16GB minimum) to run 2 databases containers plus a Quantized 7B Small Language Model.
- No GPU is required. Will strictly be using CPU only.
- Windows Users should use a VM or Cloud Instance
- If you opt for this, you must provide your own instances
- Python Version: Using Only 3.10
- (HIGHLY Recommended) Huggingface developer token saved to
HF_TOKENenvironment variable. - Install Podman (or equivalent container runtime).
- Will use OpenSearch and Neo4j containers.
LLM / Model to Pre-Download :
- Qwen2.5-7B-Instruct-1M-GGUF – Specifically the Q5_K_M version
- On the HuggingFace page, navigate to “files” and download this model.
- Drop the file into your ~/models folder. (You might need to create this.)
- Pre-pull the following containers:
- podman pull opensearchproject/opensearch:3.5.0
- podman pull opensearchproject/opensearch-dashboards:3.5.0
- podman pull neo4j:5.26.16
This workshop will only work with Python version 3.10. I would highly recommend using miniconda, uv, or venv so you can run Python 3.10 in its own isolated environment.
To install the software dependencies:
# if you are using: miniconda or venv
pip install -r requirements.txt
# OR, if using uv
uv pip install -r requirements.txt Then:
# For miniconda or venv:
pip install https://github.com/explosion/spacy-models/releases/download/en_core_web_sm-3.5.0/en_core_web_sm-3.5.0.tar.gz
# OR, if using uv:
uv pip install https://github.com/explosion/spacy-models/releases/download/en_core_web_sm-3.5.0/en_core_web_sm-3.5.0.tar.gzDone!
The workshop will consist of 4 sections. Each section will have a series of hands-on tasks for you to complete.
-
Section 1: The "R" In RAG: Use Cases & Methods
- Part 1: Simple Vector-based Agent
- Part 2: Simple BM25-based Agent
- Part 3: Simple Graph-based Agent
-
Section 2: Data Retrieval Challenges
- Part 1: Exploring Vector Retrieval
- Part 2: Exploring BM25 Retrieval
- Part 3: Exploring Graph Retrieval
-
Section 3: What Is Hybrid RAG?
- Part 1: Decouple Data & Inference
-
Section 4: Fact-based Answers Wrapped in AI Governance
- Part 1: Answers with Governance